SyVCO-RSYv5: Rolling Shutter Based Deep Learning Model for Obstacle Detection and Distance Measurement for Low-Flying Drones

Vision-based detection technologies have been raised as a useful technology for analyzing the surroundings to ensure overall safety. The obstacle detection enhances the precision of navigation and reduces collisions in low-flying drones. Conventional autonomous detection models are susceptible to various challenges, including complexities, parameter extraction, and lighting conditions. Therefore, in this research, the Synergic and Vigilance Cuculidae Optimized–Rolling Shutter based You Only Look Once-V5 (SyVCO-RSYv5) model is developed, which overcomes the real-time complexities and amplifies the detection outcome. Specifically, the optimal selection of model parameters using the SyVCO algorithm significantly enhances the performance and stability of the system, thus better adapting the proposed model to complex scenarios. Further, the incorporation of rolling shutter-based distance measurement accurately computes the distance of the detected object. Besides, the denoising performance is enhanced with the optimized vision transformer, which improves the detection performance. Further, the Multimodal Residual Invariant Feature Mapping (MRIM) method provides a robust representation of images for analyzing the obstacles in complex scenarios. When compared with other methods, the results show the enhanced efficacy of the system with an accuracy of 96.61%, specificity of 97.40%, and sensitivity of 94.80% with 80% training.